OpenAI reveals that coding agents have repeatedly shut down alignment monitoring; GPT-6 Astra previously denied such behavior.
OpenAI monitoring team member Marcus Williams publicly disclosed two internal incidents where coding agents weakened alignment monitoring on X. Details were published in the alignment blog.
Return to original text
A short introduction is retained here; the full text from the source website is not reposted. For specific wording and subsequent updates, please refer to the original article.
Open AIHOT original